xref: /freebsd/sys/ofed/drivers/infiniband/ulp/ipoib/ipoib_multicast.c (revision d6eb98610fa65663bf0df4574b7cb2c5c4ffda71)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3  *
4  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
5  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
6  * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
7  *
8  * This software is available to you under a choice of one of two
9  * licenses.  You may choose to be licensed under the terms of the GNU
10  * General Public License (GPL) Version 2, available from the file
11  * COPYING in the main directory of this source tree, or the
12  * OpenIB.org BSD license below:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
22  *      - Redistributions in binary form must reproduce the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer in the documentation and/or other materials
25  *        provided with the distribution.
26  *
27  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34  * SOFTWARE.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "ipoib.h"
41 
42 #include <linux/delay.h>
43 #include <linux/completion.h>
44 
45 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
46 static int mcast_debug_level = 1;
47 
48 module_param(mcast_debug_level, int, 0644);
49 MODULE_PARM_DESC(mcast_debug_level,
50 		 "Enable multicast debug tracing if > 0");
51 #endif
52 
53 static DEFINE_MUTEX(mcast_mutex);
54 
55 struct ipoib_mcast_iter {
56 	struct ipoib_dev_priv *priv;
57 	union ib_gid       mgid;
58 	unsigned long      created;
59 	unsigned int       queuelen;
60 	unsigned int       complete;
61 	unsigned int       send_only;
62 };
63 
64 static void ipoib_mcast_free(struct ipoib_mcast *mcast)
65 {
66 	struct ifnet *dev = mcast->priv->dev;
67 	int tx_dropped = 0;
68 
69 	ipoib_dbg_mcast(mcast->priv, "deleting multicast group %16D\n",
70 			mcast->mcmember.mgid.raw, ":");
71 
72 	if (mcast->ah)
73 		ipoib_put_ah(mcast->ah);
74 
75 	tx_dropped = mcast->pkt_queue.ifq_len;
76 	_IF_DRAIN(&mcast->pkt_queue);	/* XXX Locking. */
77 
78 	if_inc_counter(dev, IFCOUNTER_OERRORS, tx_dropped);
79 
80 	kfree(mcast);
81 }
82 
83 static struct ipoib_mcast *ipoib_mcast_alloc(struct ipoib_dev_priv *priv,
84 					     int can_sleep)
85 {
86 	struct ipoib_mcast *mcast;
87 
88 	mcast = kzalloc(sizeof *mcast, can_sleep ? GFP_KERNEL : GFP_ATOMIC);
89 	if (!mcast)
90 		return NULL;
91 
92 	mcast->priv = priv;
93 	mcast->created = jiffies;
94 	mcast->backoff = 1;
95 
96 	INIT_LIST_HEAD(&mcast->list);
97 	bzero(&mcast->pkt_queue, sizeof(mcast->pkt_queue));
98 
99 	return mcast;
100 }
101 
102 static struct ipoib_mcast *__ipoib_mcast_find(struct ipoib_dev_priv *priv,
103     void *mgid)
104 {
105 	struct rb_node *n = priv->multicast_tree.rb_node;
106 
107 	while (n) {
108 		struct ipoib_mcast *mcast;
109 		int ret;
110 
111 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
112 
113 		ret = memcmp(mgid, mcast->mcmember.mgid.raw,
114 			     sizeof (union ib_gid));
115 		if (ret < 0)
116 			n = n->rb_left;
117 		else if (ret > 0)
118 			n = n->rb_right;
119 		else
120 			return mcast;
121 	}
122 
123 	return NULL;
124 }
125 
126 static int __ipoib_mcast_add(struct ipoib_dev_priv *priv,
127     struct ipoib_mcast *mcast)
128 {
129 	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
130 
131 	while (*n) {
132 		struct ipoib_mcast *tmcast;
133 		int ret;
134 
135 		pn = *n;
136 		tmcast = rb_entry(pn, struct ipoib_mcast, rb_node);
137 
138 		ret = memcmp(mcast->mcmember.mgid.raw, tmcast->mcmember.mgid.raw,
139 			     sizeof (union ib_gid));
140 		if (ret < 0)
141 			n = &pn->rb_left;
142 		else if (ret > 0)
143 			n = &pn->rb_right;
144 		else
145 			return -EEXIST;
146 	}
147 
148 	rb_link_node(&mcast->rb_node, pn, n);
149 	rb_insert_color(&mcast->rb_node, &priv->multicast_tree);
150 
151 	return 0;
152 }
153 
154 static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
155 				   struct ib_sa_mcmember_rec *mcmember)
156 {
157 	struct ipoib_dev_priv *priv = mcast->priv;
158 	struct ifnet *dev = priv->dev;
159 	struct ipoib_ah *ah;
160 	int ret;
161 	int set_qkey = 0;
162 
163 	mcast->mcmember = *mcmember;
164 
165 	/* Set the cached Q_Key before we attach if it's the broadcast group */
166 	if (!memcmp(mcast->mcmember.mgid.raw, dev->if_broadcastaddr + 4,
167 		    sizeof (union ib_gid))) {
168 		spin_lock_irq(&priv->lock);
169 		if (!priv->broadcast) {
170 			spin_unlock_irq(&priv->lock);
171 			return -EAGAIN;
172 		}
173 		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
174 		spin_unlock_irq(&priv->lock);
175 		priv->tx_wr.remote_qkey = priv->qkey;
176 		set_qkey = 1;
177 	}
178 
179 	if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
180 		if (test_and_set_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
181 			ipoib_warn(priv, "multicast group %16D already attached\n",
182 				   mcast->mcmember.mgid.raw, ":");
183 
184 			return 0;
185 		}
186 
187 		ret = ipoib_mcast_attach(priv, be16_to_cpu(mcast->mcmember.mlid),
188 					 &mcast->mcmember.mgid, set_qkey);
189 		if (ret < 0) {
190 			ipoib_warn(priv, "couldn't attach QP to multicast group %16D\n",
191 				   mcast->mcmember.mgid.raw, ":");
192 
193 			clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags);
194 			return ret;
195 		}
196 	}
197 
198 	{
199 		struct ib_ah_attr av = {
200 			.dlid	       = be16_to_cpu(mcast->mcmember.mlid),
201 			.port_num      = priv->port,
202 			.sl	       = mcast->mcmember.sl,
203 			.ah_flags      = IB_AH_GRH,
204 			.static_rate   = mcast->mcmember.rate,
205 			.grh	       = {
206 				.flow_label    = be32_to_cpu(mcast->mcmember.flow_label),
207 				.hop_limit     = mcast->mcmember.hop_limit,
208 				.sgid_index    = 0,
209 				.traffic_class = mcast->mcmember.traffic_class
210 			}
211 		};
212 		av.grh.dgid = mcast->mcmember.mgid;
213 
214 		ah = ipoib_create_ah(priv, priv->pd, &av);
215 		if (!ah) {
216 			ipoib_warn(priv, "ib_address_create failed\n");
217 		} else {
218 			spin_lock_irq(&priv->lock);
219 			mcast->ah = ah;
220 			spin_unlock_irq(&priv->lock);
221 
222 			ipoib_dbg_mcast(priv, "MGID %16D AV %p, LID 0x%04x, SL %d\n",
223 					mcast->mcmember.mgid.raw, ":",
224 					mcast->ah->ah,
225 					be16_to_cpu(mcast->mcmember.mlid),
226 					mcast->mcmember.sl);
227 		}
228 	}
229 
230 	/* actually send any queued packets */
231 	while (mcast->pkt_queue.ifq_len) {
232 		struct mbuf *mb;
233 		_IF_DEQUEUE(&mcast->pkt_queue, mb);
234 		mb->m_pkthdr.rcvif = dev;
235 
236 		if (dev->if_transmit(dev, mb))
237 			ipoib_warn(priv, "dev_queue_xmit failed to requeue packet\n");
238 	}
239 
240 	return 0;
241 }
242 
243 static int
244 ipoib_mcast_sendonly_join_complete(int status,
245 				   struct ib_sa_multicast *multicast)
246 {
247 	struct ipoib_mcast *mcast = multicast->context;
248 	struct ipoib_dev_priv *priv = mcast->priv;
249 
250 	/* We trap for port events ourselves. */
251 	if (status == -ENETRESET)
252 		return 0;
253 
254 	if (!status)
255 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
256 
257 	if (status) {
258 		if (mcast->logcount++ < 20)
259 			ipoib_dbg_mcast(priv, "multicast join failed for %16D, status %d\n",
260 					mcast->mcmember.mgid.raw, ":", status);
261 
262 		/* Flush out any queued packets */
263 		if_inc_counter(priv->dev, IFCOUNTER_OERRORS, mcast->pkt_queue.ifq_len);
264 		_IF_DRAIN(&mcast->pkt_queue);
265 
266 		/* Clear the busy flag so we try again */
267 		status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY,
268 					    &mcast->flags);
269 	}
270 	return status;
271 }
272 
273 static int ipoib_mcast_sendonly_join(struct ipoib_mcast *mcast)
274 {
275 	struct ipoib_dev_priv *priv = mcast->priv;
276 	struct ib_sa_mcmember_rec rec = {
277 #if 0				/* Some SMs don't support send-only yet */
278 		.join_state = 4
279 #else
280 		.join_state = 1
281 #endif
282 	};
283 	int ret = 0;
284 
285 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
286 		ipoib_dbg_mcast(priv, "device shutting down, no multicast joins\n");
287 		return -ENODEV;
288 	}
289 
290 	if (test_and_set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)) {
291 		ipoib_dbg_mcast(priv, "multicast entry busy, skipping\n");
292 		return -EBUSY;
293 	}
294 
295 	rec.mgid     = mcast->mcmember.mgid;
296 	rec.port_gid = priv->local_gid;
297 	rec.pkey     = cpu_to_be16(priv->pkey);
298 
299 	mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca,
300 					 priv->port, &rec,
301 					 IB_SA_MCMEMBER_REC_MGID	|
302 					 IB_SA_MCMEMBER_REC_PORT_GID	|
303 					 IB_SA_MCMEMBER_REC_PKEY	|
304 					 IB_SA_MCMEMBER_REC_JOIN_STATE,
305 					 GFP_ATOMIC,
306 					 ipoib_mcast_sendonly_join_complete,
307 					 mcast);
308 	if (IS_ERR(mcast->mc)) {
309 		ret = PTR_ERR(mcast->mc);
310 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
311 		ipoib_warn(priv, "ib_sa_join_multicast failed (ret = %d)\n",
312 			   ret);
313 	} else {
314 		ipoib_dbg_mcast(priv, "no multicast record for %16D, starting join\n",
315 				mcast->mcmember.mgid.raw, ":");
316 	}
317 
318 	return ret;
319 }
320 
321 void ipoib_mcast_carrier_on_task(struct work_struct *work)
322 {
323 	struct ipoib_dev_priv *priv = container_of(work, struct ipoib_dev_priv,
324 						   carrier_on_task);
325 	struct ib_port_attr attr;
326 
327 	/*
328 	 * Take rtnl_lock to avoid racing with ipoib_stop() and
329 	 * turning the carrier back on while a device is being
330 	 * removed.
331 	 */
332 	if (ib_query_port(priv->ca, priv->port, &attr) ||
333 	    attr.state != IB_PORT_ACTIVE) {
334 		ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
335 		return;
336 	}
337 	if_link_state_change(priv->dev, LINK_STATE_UP);
338 }
339 
340 static int ipoib_mcast_join_complete(int status,
341 				     struct ib_sa_multicast *multicast)
342 {
343 	struct ipoib_mcast *mcast = multicast->context;
344 	struct ipoib_dev_priv *priv = mcast->priv;
345 
346 	ipoib_dbg_mcast(priv, "join completion for %16D (status %d)\n",
347 			mcast->mcmember.mgid.raw, ":", status);
348 
349 	/* We trap for port events ourselves. */
350 	if (status == -ENETRESET)
351 		return 0;
352 
353 	if (!status)
354 		status = ipoib_mcast_join_finish(mcast, &multicast->rec);
355 
356 	if (!status) {
357 		mcast->backoff = 1;
358 		mutex_lock(&mcast_mutex);
359 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
360 			queue_delayed_work(ipoib_workqueue,
361 					   &priv->mcast_task, 0);
362 		mutex_unlock(&mcast_mutex);
363 
364 		/*
365 		 * Defer carrier on work to ipoib_workqueue to avoid a
366 		 * deadlock on rtnl_lock here.
367 		 */
368 		if (mcast == priv->broadcast)
369 			queue_work(ipoib_workqueue, &priv->carrier_on_task);
370 
371 		return 0;
372 	}
373 
374 	if (mcast->logcount++ < 20) {
375 		if (status == -ETIMEDOUT || status == -EAGAIN) {
376 			ipoib_dbg_mcast(priv, "multicast join failed for %16D, status %d\n",
377 					mcast->mcmember.mgid.raw, ":", status);
378 		} else {
379 			ipoib_warn(priv, "multicast join failed for %16D, status %d\n",
380 				   mcast->mcmember.mgid.raw, ":", status);
381 		}
382 	}
383 
384 	mcast->backoff *= 2;
385 	if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
386 		mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
387 
388 	/* Clear the busy flag so we try again */
389 	status = test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
390 
391 	mutex_lock(&mcast_mutex);
392 	spin_lock_irq(&priv->lock);
393 	if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
394 		queue_delayed_work(ipoib_workqueue, &priv->mcast_task,
395 				   mcast->backoff * HZ);
396 	spin_unlock_irq(&priv->lock);
397 	mutex_unlock(&mcast_mutex);
398 
399 	return status;
400 }
401 
402 static void ipoib_mcast_join(struct ipoib_dev_priv *priv,
403     struct ipoib_mcast *mcast, int create)
404 {
405 	struct ib_sa_mcmember_rec rec = {
406 		.join_state = 1
407 	};
408 	ib_sa_comp_mask comp_mask;
409 	int ret = 0;
410 
411 	ipoib_dbg_mcast(priv, "joining MGID %16D\n",
412 	    mcast->mcmember.mgid.raw, ":");
413 
414 	rec.mgid     = mcast->mcmember.mgid;
415 	rec.port_gid = priv->local_gid;
416 	rec.pkey     = cpu_to_be16(priv->pkey);
417 
418 	comp_mask =
419 		IB_SA_MCMEMBER_REC_MGID		|
420 		IB_SA_MCMEMBER_REC_PORT_GID	|
421 		IB_SA_MCMEMBER_REC_PKEY		|
422 		IB_SA_MCMEMBER_REC_JOIN_STATE;
423 
424 	if (create) {
425 		comp_mask |=
426 			IB_SA_MCMEMBER_REC_QKEY			|
427 			IB_SA_MCMEMBER_REC_MTU_SELECTOR		|
428 			IB_SA_MCMEMBER_REC_MTU			|
429 			IB_SA_MCMEMBER_REC_TRAFFIC_CLASS	|
430 			IB_SA_MCMEMBER_REC_RATE_SELECTOR	|
431 			IB_SA_MCMEMBER_REC_RATE			|
432 			IB_SA_MCMEMBER_REC_SL			|
433 			IB_SA_MCMEMBER_REC_FLOW_LABEL		|
434 			IB_SA_MCMEMBER_REC_HOP_LIMIT;
435 
436 		rec.qkey	  = priv->broadcast->mcmember.qkey;
437 		rec.mtu_selector  = IB_SA_EQ;
438 		rec.mtu		  = priv->broadcast->mcmember.mtu;
439 		rec.traffic_class = priv->broadcast->mcmember.traffic_class;
440 		rec.rate_selector = IB_SA_EQ;
441 		rec.rate	  = priv->broadcast->mcmember.rate;
442 		rec.sl		  = priv->broadcast->mcmember.sl;
443 		rec.flow_label	  = priv->broadcast->mcmember.flow_label;
444 		rec.hop_limit	  = priv->broadcast->mcmember.hop_limit;
445 	}
446 
447 	set_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
448 	mcast->mc = ib_sa_join_multicast(&ipoib_sa_client, priv->ca, priv->port,
449 					 &rec, comp_mask, GFP_KERNEL,
450 					 ipoib_mcast_join_complete, mcast);
451 	if (IS_ERR(mcast->mc)) {
452 		clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags);
453 		ret = PTR_ERR(mcast->mc);
454 		ipoib_warn(priv, "ib_sa_join_multicast failed, status %d\n", ret);
455 
456 		mcast->backoff *= 2;
457 		if (mcast->backoff > IPOIB_MAX_BACKOFF_SECONDS)
458 			mcast->backoff = IPOIB_MAX_BACKOFF_SECONDS;
459 
460 		mutex_lock(&mcast_mutex);
461 		if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
462 			queue_delayed_work(ipoib_workqueue,
463 					   &priv->mcast_task,
464 					   mcast->backoff * HZ);
465 		mutex_unlock(&mcast_mutex);
466 	}
467 }
468 
469 void ipoib_mcast_join_task(struct work_struct *work)
470 {
471 	struct ipoib_dev_priv *priv =
472 		container_of(work, struct ipoib_dev_priv, mcast_task.work);
473 	struct ifnet *dev = priv->dev;
474 	struct ib_port_attr attr;
475 
476 	ipoib_dbg_mcast(priv, "Running join task. flags 0x%lX\n", priv->flags);
477 
478 	if (!test_bit(IPOIB_MCAST_RUN, &priv->flags))
479 		return;
480 
481 	if (ib_query_port(priv->ca, priv->port, &attr) ||
482             attr.state != IB_PORT_ACTIVE) {
483 		ipoib_dbg(priv, "%s: port state is not ACTIVE (state = %d) suspend task.\n",
484                           __func__, attr.state);
485 		return;
486 	}
487 
488 	if (ib_query_gid(priv->ca, priv->port, 0, &priv->local_gid, NULL))
489 		ipoib_warn(priv, "ib_query_gid() failed\n");
490 	else
491 		memcpy(IF_LLADDR(dev) + 4, priv->local_gid.raw, sizeof (union ib_gid));
492 
493 	{
494 		struct ib_port_attr attr;
495 
496 		if (!ib_query_port(priv->ca, priv->port, &attr))
497 			priv->local_lid = attr.lid;
498 		else
499 			ipoib_warn(priv, "ib_query_port failed\n");
500 	}
501 
502 	if (!priv->broadcast) {
503 		struct ipoib_mcast *broadcast;
504 
505 		if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
506 			return;
507 
508 		broadcast = ipoib_mcast_alloc(priv, 1);
509 		if (!broadcast) {
510 			ipoib_warn(priv, "failed to allocate broadcast group\n");
511 			mutex_lock(&mcast_mutex);
512 			if (test_bit(IPOIB_MCAST_RUN, &priv->flags))
513 				queue_delayed_work(ipoib_workqueue,
514 						   &priv->mcast_task, HZ);
515 			mutex_unlock(&mcast_mutex);
516 			return;
517 		}
518 
519 		spin_lock_irq(&priv->lock);
520 		memcpy(broadcast->mcmember.mgid.raw, dev->if_broadcastaddr + 4,
521 		       sizeof (union ib_gid));
522 		priv->broadcast = broadcast;
523 
524 		__ipoib_mcast_add(priv, priv->broadcast);
525 		spin_unlock_irq(&priv->lock);
526 	}
527 
528 	if (priv->broadcast &&
529 	    !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
530 		if (priv->broadcast &&
531 		    !test_bit(IPOIB_MCAST_FLAG_BUSY, &priv->broadcast->flags))
532 			ipoib_mcast_join(priv, priv->broadcast, 0);
533 		return;
534 	}
535 
536 	while (1) {
537 		struct ipoib_mcast *mcast = NULL;
538 
539 		spin_lock_irq(&priv->lock);
540 		list_for_each_entry(mcast, &priv->multicast_list, list) {
541 			if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)
542 			    && !test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags)
543 			    && !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
544 				/* Found the next unjoined group */
545 				break;
546 			}
547 		}
548 		spin_unlock_irq(&priv->lock);
549 
550 		if (&mcast->list == &priv->multicast_list) {
551 			/* All done */
552 			break;
553 		}
554 
555 		ipoib_mcast_join(priv, mcast, 1);
556 		return;
557 	}
558 
559 	spin_lock_irq(&priv->lock);
560 	if (priv->broadcast)
561 		priv->mcast_mtu = IPOIB_UD_MTU(ib_mtu_enum_to_int(priv->broadcast->mcmember.mtu));
562 	else
563 		priv->mcast_mtu = priv->admin_mtu;
564 	spin_unlock_irq(&priv->lock);
565 
566 	if (!ipoib_cm_admin_enabled(priv))
567 		ipoib_change_mtu(priv, min(priv->mcast_mtu, priv->admin_mtu),
568 		    true);
569 
570 	ipoib_dbg_mcast(priv, "successfully joined all multicast groups\n");
571 
572 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
573 }
574 
575 int ipoib_mcast_start_thread(struct ipoib_dev_priv *priv)
576 {
577 	ipoib_dbg_mcast(priv, "starting multicast thread flags 0x%lX\n",
578 	    priv->flags);
579 
580 	mutex_lock(&mcast_mutex);
581 	if (!test_and_set_bit(IPOIB_MCAST_RUN, &priv->flags))
582 		queue_delayed_work(ipoib_workqueue, &priv->mcast_task, 0);
583 	mutex_unlock(&mcast_mutex);
584 
585 	return 0;
586 }
587 
588 int ipoib_mcast_stop_thread(struct ipoib_dev_priv *priv, int flush)
589 {
590 
591 	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
592 
593 	mutex_lock(&mcast_mutex);
594 	clear_bit(IPOIB_MCAST_RUN, &priv->flags);
595 	cancel_delayed_work(&priv->mcast_task);
596 	mutex_unlock(&mcast_mutex);
597 
598 	if (flush)
599 		flush_workqueue(ipoib_workqueue);
600 
601 	return 0;
602 }
603 
604 static int ipoib_mcast_leave(struct ipoib_dev_priv *priv, struct ipoib_mcast *mcast)
605 {
606 	int ret = 0;
607 
608 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
609 		ib_sa_free_multicast(mcast->mc);
610 
611 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_ATTACHED, &mcast->flags)) {
612 		ipoib_dbg_mcast(priv, "leaving MGID %16D\n",
613 				mcast->mcmember.mgid.raw, ":");
614 
615 		/* Remove ourselves from the multicast group */
616 		ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
617 				      be16_to_cpu(mcast->mcmember.mlid));
618 		if (ret)
619 			ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
620 	}
621 
622 	return 0;
623 }
624 
625 void
626 ipoib_mcast_send(struct ipoib_dev_priv *priv, void *mgid, struct mbuf *mb)
627 {
628 	struct ifnet *dev = priv->dev;
629 	struct ipoib_mcast *mcast;
630 
631 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)		||
632 	    !priv->broadcast					||
633 	    !test_bit(IPOIB_MCAST_FLAG_ATTACHED, &priv->broadcast->flags)) {
634 		if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
635 		m_freem(mb);
636 		return;
637 	}
638 
639 	mcast = __ipoib_mcast_find(priv, mgid);
640 	if (!mcast) {
641 		/* Let's create a new send only group now */
642 		ipoib_dbg_mcast(priv, "setting up send only multicast group for %16D\n",
643 				mgid, ":");
644 
645 		mcast = ipoib_mcast_alloc(priv, 0);
646 		if (!mcast) {
647 			ipoib_warn(priv, "unable to allocate memory for "
648 				   "multicast structure\n");
649 			if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
650 			m_freem(mb);
651 			goto out;
652 		}
653 
654 		set_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags);
655 		memcpy(mcast->mcmember.mgid.raw, mgid, sizeof (union ib_gid));
656 		__ipoib_mcast_add(priv, mcast);
657 		list_add_tail(&mcast->list, &priv->multicast_list);
658 	}
659 
660 	if (!mcast->ah) {
661 		if (mcast->pkt_queue.ifq_len < IPOIB_MAX_MCAST_QUEUE) {
662 			_IF_ENQUEUE(&mcast->pkt_queue, mb);
663 		} else {
664 			if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
665 			m_freem(mb);
666 		}
667 
668 		if (test_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
669 			ipoib_dbg_mcast(priv, "no address vector, "
670 					"but multicast join already started\n");
671 		else if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
672 			ipoib_mcast_sendonly_join(mcast);
673 
674 		/*
675 		 * If lookup completes between here and out:, don't
676 		 * want to send packet twice.
677 		 */
678 		mcast = NULL;
679 	}
680 
681 out:
682 	if (mcast && mcast->ah)
683 		ipoib_send(priv, mb, mcast->ah, IB_MULTICAST_QPN);
684 }
685 
686 void ipoib_mcast_dev_flush(struct ipoib_dev_priv *priv)
687 {
688 	LIST_HEAD(remove_list);
689 	struct ipoib_mcast *mcast, *tmcast;
690 	unsigned long flags;
691 
692 	ipoib_dbg_mcast(priv, "flushing multicast list\n");
693 
694 	spin_lock_irqsave(&priv->lock, flags);
695 
696 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
697 		list_del(&mcast->list);
698 		rb_erase(&mcast->rb_node, &priv->multicast_tree);
699 		list_add_tail(&mcast->list, &remove_list);
700 	}
701 
702 	if (priv->broadcast) {
703 		rb_erase(&priv->broadcast->rb_node, &priv->multicast_tree);
704 		list_add_tail(&priv->broadcast->list, &remove_list);
705 		priv->broadcast = NULL;
706 	}
707 
708 	spin_unlock_irqrestore(&priv->lock, flags);
709 
710 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
711 		ipoib_mcast_leave(priv, mcast);
712 		ipoib_mcast_free(mcast);
713 	}
714 }
715 
716 static int ipoib_mcast_addr_is_valid(const u8 *addr, unsigned int addrlen,
717 				     const u8 *broadcast)
718 {
719 	if (addrlen != INFINIBAND_ALEN)
720 		return 0;
721 	/* reserved QPN, prefix, scope */
722 	if (memcmp(addr, broadcast, 6))
723 		return 0;
724 	/* signature lower, pkey */
725 	if (memcmp(addr + 7, broadcast + 7, 3))
726 		return 0;
727 	return 1;
728 }
729 
730 void ipoib_mcast_restart_task(struct work_struct *work)
731 {
732 	struct ipoib_dev_priv *priv =
733 		container_of(work, struct ipoib_dev_priv, restart_task);
734 	ipoib_mcast_restart(priv);
735 }
736 
737 void ipoib_mcast_restart(struct ipoib_dev_priv *priv)
738 {
739 	struct ifnet *dev = priv->dev;
740 	struct ifmultiaddr *ifma;
741 	struct ipoib_mcast *mcast, *tmcast;
742 	LIST_HEAD(remove_list);
743 	struct ib_sa_mcmember_rec rec;
744 	int addrlen;
745 
746 	ipoib_dbg_mcast(priv, "restarting multicast task flags 0x%lX\n",
747 	    priv->flags);
748 
749 	ipoib_mcast_stop_thread(priv, 0);
750 
751 	if_maddr_rlock(dev);
752 	spin_lock(&priv->lock);
753 
754 	/*
755 	 * Unfortunately, the networking core only gives us a list of all of
756 	 * the multicast hardware addresses. We need to figure out which ones
757 	 * are new and which ones have been removed
758 	 */
759 
760 	/* Clear out the found flag */
761 	list_for_each_entry(mcast, &priv->multicast_list, list)
762 		clear_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
763 
764 	/* Mark all of the entries that are found or don't exist */
765 
766 
767 	CK_STAILQ_FOREACH(ifma, &dev->if_multiaddrs, ifma_link) {
768 		union ib_gid mgid;
769 		uint8_t *addr;
770 
771 		if (ifma->ifma_addr->sa_family != AF_LINK)
772 			continue;
773 		addr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
774 		addrlen = ((struct sockaddr_dl *)ifma->ifma_addr)->sdl_alen;
775 		if (!ipoib_mcast_addr_is_valid(addr, addrlen,
776 					       dev->if_broadcastaddr))
777 			continue;
778 
779 		memcpy(mgid.raw, addr + 4, sizeof mgid);
780 
781 		mcast = __ipoib_mcast_find(priv, &mgid);
782 		if (!mcast || test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
783 			struct ipoib_mcast *nmcast;
784 
785 			/* ignore group which is directly joined by userspace */
786 			if (test_bit(IPOIB_FLAG_UMCAST, &priv->flags) &&
787 			    !ib_sa_get_mcmember_rec(priv->ca, priv->port, &mgid, &rec)) {
788 				ipoib_dbg_mcast(priv, "ignoring multicast entry for mgid %16D\n",
789 						mgid.raw, ":");
790 				continue;
791 			}
792 
793 			/* Not found or send-only group, let's add a new entry */
794 			ipoib_dbg_mcast(priv, "adding multicast entry for mgid %16D\n",
795 					mgid.raw, ":");
796 
797 			nmcast = ipoib_mcast_alloc(priv, 0);
798 			if (!nmcast) {
799 				ipoib_warn(priv, "unable to allocate memory for multicast structure\n");
800 				continue;
801 			}
802 
803 			set_bit(IPOIB_MCAST_FLAG_FOUND, &nmcast->flags);
804 
805 			nmcast->mcmember.mgid = mgid;
806 
807 			if (mcast) {
808 				/* Destroy the send only entry */
809 				list_move_tail(&mcast->list, &remove_list);
810 
811 				rb_replace_node(&mcast->rb_node,
812 						&nmcast->rb_node,
813 						&priv->multicast_tree);
814 			} else
815 				__ipoib_mcast_add(priv, nmcast);
816 
817 			list_add_tail(&nmcast->list, &priv->multicast_list);
818 		}
819 
820 		if (mcast)
821 			set_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags);
822 	}
823 
824 	/* Remove all of the entries don't exist anymore */
825 	list_for_each_entry_safe(mcast, tmcast, &priv->multicast_list, list) {
826 		if (!test_bit(IPOIB_MCAST_FLAG_FOUND, &mcast->flags) &&
827 		    !test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags)) {
828 			ipoib_dbg_mcast(priv, "deleting multicast group %16D\n",
829 					mcast->mcmember.mgid.raw, ":");
830 
831 			rb_erase(&mcast->rb_node, &priv->multicast_tree);
832 
833 			/* Move to the remove list */
834 			list_move_tail(&mcast->list, &remove_list);
835 		}
836 	}
837 
838 	spin_unlock(&priv->lock);
839 	if_maddr_runlock(dev);
840 
841 	/* We have to cancel outside of the spinlock */
842 	list_for_each_entry_safe(mcast, tmcast, &remove_list, list) {
843 		ipoib_mcast_leave(mcast->priv, mcast);
844 		ipoib_mcast_free(mcast);
845 	}
846 
847 	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
848 		ipoib_mcast_start_thread(priv);
849 }
850 
851 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
852 
853 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct ipoib_dev_priv *priv)
854 {
855 	struct ipoib_mcast_iter *iter;
856 
857 	iter = kmalloc(sizeof *iter, GFP_KERNEL);
858 	if (!iter)
859 		return NULL;
860 
861 	iter->priv = priv;
862 	memset(iter->mgid.raw, 0, 16);
863 
864 	if (ipoib_mcast_iter_next(iter)) {
865 		kfree(iter);
866 		return NULL;
867 	}
868 
869 	return iter;
870 }
871 
872 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
873 {
874 	struct ipoib_dev_priv *priv = iter->priv;
875 	struct rb_node *n;
876 	struct ipoib_mcast *mcast;
877 	int ret = 1;
878 
879 	spin_lock_irq(&priv->lock);
880 
881 	n = rb_first(&priv->multicast_tree);
882 
883 	while (n) {
884 		mcast = rb_entry(n, struct ipoib_mcast, rb_node);
885 
886 		if (memcmp(iter->mgid.raw, mcast->mcmember.mgid.raw,
887 			   sizeof (union ib_gid)) < 0) {
888 			iter->mgid      = mcast->mcmember.mgid;
889 			iter->created   = mcast->created;
890 			iter->queuelen  = mcast->pkt_queue.ifq_len;
891 			iter->complete  = !!mcast->ah;
892 			iter->send_only = !!(mcast->flags & (1 << IPOIB_MCAST_FLAG_SENDONLY));
893 
894 			ret = 0;
895 
896 			break;
897 		}
898 
899 		n = rb_next(n);
900 	}
901 
902 	spin_unlock_irq(&priv->lock);
903 
904 	return ret;
905 }
906 
907 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
908 			   union ib_gid *mgid,
909 			   unsigned long *created,
910 			   unsigned int *queuelen,
911 			   unsigned int *complete,
912 			   unsigned int *send_only)
913 {
914 	*mgid      = iter->mgid;
915 	*created   = iter->created;
916 	*queuelen  = iter->queuelen;
917 	*complete  = iter->complete;
918 	*send_only = iter->send_only;
919 }
920 
921 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
922